{
  "abstract": "Background Accurately predicting microvascular invasion (MVI) preoperatively remains a core challenge for precision surgical decision-making in patients with colorectal liver metastases (CRLM). This study developed and validated a three-dimensional multimodal fractal dynamic model (FRAC-3D) that integrates spatial metabolism, cytoarchitecture, and blood perfusion, aiming to non-invasively identify MVI status preoperatively and evaluate patient prognosis. Methods Utilizing a retrospective nested cohort design, stabilized inverse probability of treatment weighting (sIPTW) was applied to the derivation cohort (n = 100) for baseline debiasing. A core cohort of 145 patients with complete PET/CT, CE-MRI, and DWI imaging was ultimately included. Spatial metabolic fractal dimension (FD_PET), architectural fractal dimension (FD_ADC), and perfusion fractal acceleration (Afd_MRI) were extracted to construct the FRAC-3D model. The model’s discrimination, calibration, and clinical benefit were validated in an independent temporal validation cohort (n = 45). Results Multivariable analysis confirmed that FD_ADC is the strongest independent predictor of MVI (OR = 3.85, 95% CI: 2.45–6.04, P < 0.001). The FRAC-3D model achieved an AUC of 0.938 and a negative predictive value of 96.2% in the validation cohort. Compared with the dual-modal model, FRAC-3D significantly improved risk restratification (continuous NRI = 0.425, P < 0.001). A “highly chaotic” imaging phenotype, identified based on fractal feature clustering, demonstrated an MVI incidence as high as 96.7% and a median recurrence-free survival (RFS) of only 5.2 months. Conclusion Through the cross-validation of multidimensional biophysical features, the FRAC-3D model enables high-precision preoperative prediction of MVI status. This study advances the perioperative assessment of CRLM from traditional ‘static anatomical metrics’ into a novel dimension of ‘microenvironmental dynamic evolution’, providing a robust, physics-driven tool for individualized surgical margin planning and the precise stratification of targeted neoadjuvant therapy.",
  "abstract_shown": true,
  "affiliations": [
    "Department of Nuclear Medicine, First Affiliated Hospital of Jinzhou Medical University",
    "Architecture and Civil Engineering Institute, Guangdong University of Petrochemical Technology",
    "Department of Radiology, First Affiliated Hospital of Jinzhou Medical University"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 5,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "Ziyi Guo"
    },
    {
      "affs": [
        1
      ],
      "name": "Chao Guo"
    },
    {
      "affs": [
        1
      ],
      "name": "Zhengran Lu"
    },
    {
      "affs": [
        2
      ],
      "name": "Kailin Zou"
    },
    {
      "affs": [
        2
      ],
      "name": "Hongnan Zhang"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/9v75k4t6wknsd.bib",
  "cited_by": 0,
  "doi": "10.1186/s12967-026-08712-y",
  "fulltext_url": "https://doi.org/10.1186/s12967-026-08712-y",
  "id": "9v75k4t6wknsd",
  "issns": [
    "1479-5876"
  ],
  "landing_url": "https://doi.org/10.1186/s12967-026-08712-y",
  "language": "en",
  "licence": "CC-BY",
  "licence_label": "CC BY",
  "licence_source": "journal",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "lcc",
      "R"
    ],
    [
      "openalex",
      "2741|Radiomics and Machine Learning in Medical Imaging"
    ]
  ],
  "openalex": "W7171441026",
  "pdf_url": null,
  "pmcid": null,
  "pmid": null,
  "published": "2026-07-01",
  "publisher": "BMC",
  "ris": "https://research.draconitic.com/paper/9v75k4t6wknsd.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T13:46:24Z",
      "native_id": "93fbbd17beb44210a4f9febb36221349",
      "source": "doaj",
      "url": "https://doi.org/10.1186/s12967-026-08712-y"
    }
  ],
  "subjects": [
    {
      "field": "medicine",
      "origin": "openalex:2741|Radiomics and Machine Learning in Medical Imaging",
      "subfield": "imaging"
    }
  ],
  "title": "Three-dimensional multimodal fractal dynamics integrating spatial metabolism, cytoarchitecture, and perfusion for preoperative prediction of microvascular invasion in colorectal liver metastases",
  "title_text": "Three-dimensional multimodal fractal dynamics integrating spatial metabolism, cytoarchitecture, and perfusion for preoperative prediction of microvascular invasion in colorectal liver metastases",
  "type": "peer-reviewed",
  "updated_at": "2026-10-30T13:46:24Z",
  "url": "https://research.draconitic.com/paper/9v75k4t6wknsd",
  "venue": "Journal of Translational Medicine",
  "versions": []
}